bug#81509: 31.0.91; Calling dired on nonexistent directory changes current buffer

> From: Stephen Berman <[email protected]>
> Cc: Eli Zaretskii <[email protected]>, Madhu <[email protected]>
> Date: Tue, 28 Jul 2026 15:55:29 +0200
> 
> This behavior is a by-product of 3b7d9e37ce0c.  (BTW, the nonexistent
> directory doesn't have to be represented by a wildcard in the sense of a
> shell glob (e.g. *nonexist*), i.e. C-x d /tmp/a/nonexist also shows the
> same behavior.  But /tmp/a/nonexist is a wildcard in the sense of
> insert-directory and maybe that's what the OP meant.)  Before that
> change, when attempting to visit in Dired a nonexistent directory,
> insert-directory signalled an error, which short-circuited the Dired
> command, so that the current buffer when the Dired command was invoked
> remained current.  Since 3b7d9e37ce0c, ls errors, such as trying to
> display a nonexistent directory, are now shown in a special buffer after
> completing the Dired command, as you observed.  But the Dired commands
> use `pop-to-buffer-same-window' and
> `switch-to-buffer-other-{window,frame,tab}' and calling these with a
> nonexistent directory - i.e. with nil - as argument makes another buffer
> current.  I agree this is surprising behavior.  The attached patch
> prevents this, i.e., the buffer that was current when the Dired command
> was invoked remains current when the directory to be visited is
> nonexistent.  Madhu, can you confirm?
> 
> Eli, if this patch DTRT for Madhu can it be installed in emacs-31, since
> the undesirable behavior was due to changes there?

Yes, please.